Are GAN generated images easy to detect? A critical analysis of the state-of-the-art

The advent of deep learning has brought a significant improvement in the\nquality of generated media. However, with the increased level of photorealism,\nsynthetic media are becoming hardly distinguishable from real ones, raising\nserious concerns about the spread of fake or manipulated information over the\nInternet. In this context, it is important to develop automated tools to\nreliably and timely detect synthetic media. In this work, we analyze the\nstate-of-the-art methods for the detection of synthetic images, highlighting\nthe key ingredients of the most successful approaches, and comparing their\nperformance over existing generative architectures. We will devote special\nattention to realistic and challenging scenarios, like media uploaded on social\nnetworks or generated by new and unseen architectures, analyzing the impact of\nsuitable augmentation and training strategies on the detectors' generalization\nability.\n